Spaces:
Running
Running
File size: 5,359 Bytes
4e5eb13 452c890 34660db 452c890 34660db 452c890 34660db 948cba3 452c890 4e5eb13 948cba3 09d7cf1 f36044d 09d7cf1 f36044d 09d7cf1 f36044d 09d7cf1 f36044d b471609 09d7cf1 948cba3 0123b5a 948cba3 34660db 948cba3 34660db 948cba3 34660db 948cba3 34660db 09d7cf1 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 |
import gradio as gr
import pandas as pd
# Static data - reordered columns: Method, #Param., Input Type, Control Type, Model Type, Mean Traj. ↓, Acc. ↑
STATIC_DATA = [
["w/o WM", "72B", "RGB", "–", "VLM", 6.24, 50.27],
["PathDreamer [36]", "0.69B", "RGB-D; Sem; Pano", "Viewpoint", "Image Gen.", 5.28, 56.99],
["SE3DS [11]", "1.1B", "RGB-D; Pano", "Viewpoint", "Image Gen.", 5.29, 57.53],
["NWM [25]", "1B", "RGB", "Trajectory", "Video Gen.", 5.68, 57.35],
["SVD [6]", "1.5B", "RGB", "Image", "Video Gen.", 5.29, 57.71],
["LTX-Video [5]", "2B", "RGB", "Text", "Video Gen.", 5.37, 56.08],
["Hunyuan [4]", "13B", "RGB", "Text", "Video Gen.", 5.21, 57.71],
["Wan2.1 [23]", "14B", "RGB", "Text", "Video Gen.", 5.24, 58.26],
["Cosmos [1]", "2B", "RGB", "Text", "Video Gen.", 5.898, 52.27],
["Runway", "–", "–", "Text", "Video Gen.", "–", "–"],
["SVD† [6]", "1.5B", "RGB; Pano", "Action", "Video Gen. Post-Train", 5.02, 60.98],
["LTX† [5]", "2B", "RGB; Pano", "Action", "Video Gen. Post-Train", 5.49, 57.53],
["WAN2.1† [23]", "14B", "RGB; Pano", "Action", "Video Gen. Post-Train", "XXX", "XXX"],
["Cosmos† [1]", "2B", "RGB; Pano", "Action", "Video Gen. Post-Train", 5.08, 60.25],
]
COLUMNS = ["Method", "#Param.", "Input Type", "Control Type", "Model Type", "Mean Traj. ↓", "Acc. ↑"]
def create_leaderboard():
df = pd.DataFrame(STATIC_DATA, columns=COLUMNS)
# Sort by accuracy in descending order (highest first), handling non-numeric values
df_clean = df.copy()
# Replace non-numeric values with -1 for sorting (so they appear at bottom)
df_clean['Acc. ↑'] = pd.to_numeric(df_clean['Acc. ↑'], errors='coerce').fillna(-1)
df_sorted = df_clean.sort_values('Acc. ↑', ascending=False)
# Return original df with the sorted order but original values
return df.iloc[df_sorted.index].reset_index(drop=True)
with gr.Blocks(title="World-in-World: Building a Closed-Loop World Interface to Evaluate World Models", theme=gr.themes.Soft()) as demo:
gr.HTML("<h1 style='text-align: center; margin-bottom: 1rem'>🏆 World-in-World: Building a Closed-Loop World Interface to Evaluate World Models</h1>")
with gr.Tabs():
# New Zone before Leaderboard
with gr.TabItem("🧑🏫 Instruction & Environmental Feedback"):
with gr.Row():
with gr.Column(scale=1, min_width=300): # Ensuring proper alignment
gr.HTML("<h3 style='text-align: center;'>Instruction:</h3>")
gr.Markdown("Navigate to the Toaster in the room and be as close as possible to it.")
gr.Markdown("""
**Environment Step 4-7:**
**Planning:**
1. Move leftward by 0.25.
2. Move leftward by 0.25.
3. Move forward by 0.25.
4. Move forward by 0.25.
""")
with gr.Column(scale=2, min_width=600):
gr.HTML("<h3 style='text-align: center;'>Closed-Loop Environmental Feedback</h3>")
# Adjust image/video sizes for better layout
gr.Video("/home/user/app/demo_source_data/AR/FTwan21_lora/X7HyMhZNoso/E145/A001/world_model_gen/bbox_gen_video_1.mp4", label="Left Image", width=480)
gr.Image("/home/user/app/demo_source_data/AR/FTwan21_lora/X7HyMhZNoso/E145/A001/real_obs.png", label="Bird Eye View", type="pil", width=480)
gr.HTML("<h4 style='text-align: center;'>3D Scene:</h4>")
with gr.Column(scale=1):
gr.Model3D("/home/user/app/demo_source_data/scenes_glb/5ZKStnWn8Zo.glb", label="3D Scene")
with gr.TabItem("📊 Leaderboard"):
leaderboard_table = gr.DataFrame(
value=create_leaderboard(),
headers=COLUMNS,
datatype=["str", "str", "str", "str", "str", "number", "number"],
interactive=False,
wrap=True
)
with gr.TabItem("📝 About"):
gr.Markdown("""
# World-in-World: Building a Closed-Loop World Interface to Evaluate World Models
This leaderboard showcases performance metrics across different types of AI models in world modeling tasks:
## Model Categories
- **VLM**: Vision-Language Models
- **Image Gen.**: Image Generation Models
- **Video Gen.**: Video Generation Models
- **Video Gen. Post-Train**: Post-training specialized Video Generation Models
## Metrics Explained
- **Acc. ↑**: Accuracy score (higher values indicate better performance)
- **Mean Traj. ↓**: Mean trajectory error (lower values indicate better performance)
## Notes
- † indicates post-training specialized models
- XXX indicates results pending/unavailable
- – indicates not applicable or not available
*Results represent performance on world modeling evaluation benchmarks and may vary across different evaluation settings.*
""")
if __name__ == "__main__":
demo.launch()
|